soak

Rapid, reproducible qualitative data analysis with LLMs.

Soak helps qualitative researchers rapidly define and run qualitative data analyses. It provides graph-based pipelines for thematic analysis, classification, and structured data extraction from unstructured text. It encourages expert supervision and oversight of AI analyses and facilitates transparent sharing of reproducible research.

Tutorials

• Getting Started – Installation and your first analysis
• Customizing Your Analysis – Adapting prompts and directing the LLM to answer specific research questions.
• Working with Results – Understanding codes, themes, and exports. Comparing analysis outputs.

How-to Guides

Analysis workflows

• Thematic Analysis – Inductive coding and theme generation
• Build a Classifier – Structured classification pipelines
• Ground Truth Validation – Validate against labelled data

Working with data

• Working with Spreadsheet Data – Process CSV and Excel files
• Pre-extraction Workflow – Filter text before analysis
• Adapting Pipelines – Customizing pipeline workflows

Explanations

• What is soak? – Purpose and design philosophy
• DAG Architecture – Why pipelines, execution model
• Node Types – Understanding different processing nodes
• Template System – Jinja2 and struckdown syntax
• Quote Verification – How quote validation works
• Model Aliases – Configuring models for pipeline roles

Reference

• CLI Reference – Command-line interface
• Node Reference – All node types and parameters
• Quote Verification Algorithm – Technical specification

---

Example outputs

• Thematic analysis (simple) – Analysis of patient interviews
• Thematic analysis (extended) – Same data, different model
• Analysis comparison – Comparing two analyses
• Classifier output – Structured data extraction

Support

• GitHub: github.com/benwhalley/soak
• Issues: github.com/benwhalley/soak/issues